Parties for Rent?

Ambition, Ideology, and Party

Switching in Brazils Chamber of Deputies
Scott W. Desposato

University of California, San Diego

Party switching by legislators has been common in many countries, including the Philippines, Italy, Nepal, Ecuador, Russia,
and Japan. While frequently dismissed as simply an indicator of a weak parties, switching provides a unique window on
party systems. To the extent that we understand affiliation decisions, we gain insight on the way politicians use parties to
advance their careers. In this article I offer a model of party-membership patterns, where decisions to switch party or to
stay put are a function of the strategic interaction of legislators and endogenous party leaders. I test the model on the case
of Brazil, where switching is common. Results suggest that Brazilian legislators use parties to maximize pork, ideological
consistency, and short-term electoral success, but which of these matters most depends on constituents, i.e., legislators use
parties for different purposes in different electoral environments. The approach developed here could easily be applied to
study legislative behavior in other political systems.

adisons admonishments about the dangers of

faction aside, most scholars agree that parties
are quite useful for the consolidation of effective and stable democracies. Political parties organize and
aggregate social interests, consolidating lines of political
conflict. They regularize democratic practices and serve
as mechanisms for compromise and representation. They
decrease voters information costs (especially in an environment with many candidates) and increase governments accountability to voters (Aldrich 1995; Cox and
McCubbins 1993; Snyder and Ting 2001). Scholars have
even linked party systems characteristics to countries
economic policies (Haggard and Kaufman 1995; Johnson
and Crisp 2003; Mainwaring 1999)
While parties are useful for democracies, scholars argue that parties are equally useful for ambitious politicians. During elections, parties can provide campaign
workers, financial support, and well-developed policy
brand names. Parties can control access to desired career opportunities and may enhance politicians access to
government pork.
Many countries, however, have party systems that
apparently provide few such benefits to polity or politicians. Parties in these systems are weakly institutionalized,

very fluid, and highly volatile. They typically lack distinct

ideological platforms, have low discipline within legislative delegations, and suffer volatile levels of popular support. Given the important role parties play in shaping and
stabilizing democratic political arenas, a key task for comparative political scholars has been explaining the nature
of party systems and suggesting mechanisms to strengthen
them.1
One oft-overlooked window on party systems is
switching by politicians. While switching is relatively rare
in most countries, it has been common in many countries,
including South Africa, Japan, Bolivia, Ecuador, Nepal,
Russia, the Philippines, France, Italy, and Brazil. Such behavior is usually dismissed as an indicator that parties
dont matter, but I argue that party switching warrants
study for at least three reasons.
First, frequent switching makes it clear that parties
do matterotherwise politicians would not bother to
switch. Second, and more importantly, switching provides
a unique window on politicians underlying preferences,
including their incentives for belonging to political parties. An examination of patterns of party affiliations can
reveal the roles parties play in meeting politicians varied career challenges. This increases our understanding

Scott Desposato is assistant professor of political science, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA
(swd@ucsd.edu).
For their comments and suggestions, I thank Ben Bishin, Brian Crisp, Greg Johnson, David Karol, Kris Kanthak, Laura Langer, Mark Hugo
Lopez, Bill Mishler, Pam Singh, Ben Smith, and Mike Thies. For excellent research assistance, I thank Moema Bonelli.
1

See Carey and Shugart (1995); Bowler, Farrell, and Katz (1999); Ames (2001), and National Democratic Institute (1997).

American Journal of Political Science, Vol. 50, No. 1, January 2006, Pp. 6280


C 2006,

62

Midwest Political Science Association

ISSN 0092-5853

PARTIES FOR RENT

of inchoate party systems; it also aids in the design of

party-strengthening institutions. Finally, switching poses
a normative problem for representation in mass democracies. Parties are the primary mechanism linking voters and
politicians in modern mass democracies. Meaningful and
stable party labels enable voters to make identify optimal
candidates and cast appropriate ballots. Party switching,
however, violates the basic electoral pact and effectively
makes party labels meaningless.
In this article I offer a model of party affiliations,
where membership patterns are a function of the strategic interaction of individual legislators and political parties. Legislators try to maximize utility, a function of the
payoffs of membership in each party and their own attributes, less a transaction cost associated with switching.
Parties invite, reject, or expel members to maximize the
utility of a majority of their current members, a function
of exogenous and endogenous resource endowments.
I apply my model to the case of Brazil. I show that
political institutions and mass attitudes combine to create
a very fluid market for parties and legislators, consistent
with observed frequent switching. More than a third of
deputies switch party during their terms, some as many
as seven times. Brazilian parties lack of cohesion and
stability has earned them the label of party for rent
and inspired Sartori (1993) to call Brazil, the anti-party
system.
The empirical results are consistent with theoretical
predictions and offer several insights about the Brazilian party system. The strongest motives for party affiliation are access to distributive resources, electoral opportunities, and compatible policy positions, but which is
most important varies with voters characteristics. In lessdeveloped electoral environments, legislators are most
concerned with parties access to government largess. In
more developed regions, legislators are less concerned
with pork access and relatively more concerned with ideological credibility.
The article proceeds in several steps. First, I present
a formal model of party switching and discusses its empirical implications. Then I test the model on the case of
Brazil and discuss broader implications and generalization to other cases.

A Model of Party Switching

In this section, I build a general model that predicts the
frequency and patterns of party system across different
types of political systems. Previous work, both empirical
and theoretical, has focused on building country-specific

63
explanations of party-membership patterns. I briefly review the findings from this literature, then offer a general
model that has empirical implications for all democratic
systems.
Most previous work on party switching is empirical,
examining explanations for and the impact of switching
in countries where it is frequent. In nearly all cases, however, a common theme is the intersection of institutions,
ambition, and constituency characteristics. For example,
Mershon and Heller (2004a) link frequent switching in
Italy to electoral rules, discipline, and party size, arguing that the interaction of discipline and policy disagreements prompts MPs to leave their parties. Mejia Acosta
(2004) explores switching in Ecuador, explaining variance as a function of district magnitude, party size, and
party ideology. Several scholars have explored the dynamics of the LDPs breakup in Japan. Cox and Rosenbluth
(1995) argue that the end of the Cold War and an economic downturn led to factional splits and defections,
especially by electorally marginal parliamentarians. Reed
and Scheiner (2002) focus on the behavior of individual
legislators, finding that defection from the ruling party is
predicted by policy and foward-looking institutional preferences of legislators, in addition to immediate electoral
concerns.
Two important theoretical contributions are the
work of Aldrich and Bianco (1992) and Mershon and
Heller (2004b). Both model legislators decisions as
interdependentone legislators decision depends on
the anticipated strategic behavior of other legislators.
Mershon and Heller (2004b) take an important step forward by examining the interaction of legislators and party
leaders. They model party leaders as setting levels of discipline and legislators responding with switch or stay
put strategies. They predict cascadeswhere one switch
prompts a series of party-membership changes. Their
model is entirely spatial in payoffs and utilities and does
not fully endogenize the role of party leaders.
Most relevant for this project is the work of Aldrich
and Bianco (1992), who model candidates decisions to
join one of two parties prior to an election. They examine the strategic behavior of challengers and incumbents
seeking reelection. Their model includes primary and general elections and the value of party labels. They find that
competition for a partys nomination can prompt party
switches, and in the context of declining party strength,
will lead to cascades of party defections. Their model is
built for the United States single-member district, twoparty system, and predicts switching in the context of a
decaying party or realignment (specifically the collapse of
the Whigs in the 1850s or the realignment of conservative
Southerners since the 1970s).

64

SCOTT W. DESPOSATO

TABLE 1 Typologies of Legislative Party

Systems
Payoff Type

Party Control?
No
Yes

Club

Private

I
III

II
IV

In this section I present an alternative model of

party membership. My approach is very similar to that
of Aldrich and Bianco (1992), but generalizes their approach to other political systems, allowing for different
kinds of payoffs to membership, different kinds of party
structure, and endogenous party leadership decisions. In
the following I present a typology of party systems, consider the logic of a market for party membership, then
formalize the model and derive empirical implications.
A Typology of Party Systems. I propose conceptualizing
legislative party systems according to the nature of the
benefits that party membership provides and the extent
to which parties control access to such benefits, summarized in Table 1. The benefits of party membership may be
classified either as private or as club goods. Private goods
are both rival and exclusive, meaning that their consumption reduces the amount of that good available to others,
and others can be excluded from enjoying the same good.
Examples in legislatures often include pork, committee
assignments, and nominations for elected office. Club
goods fall between private and public goods. They are
nonexclusive only for club members, all of whom automatically enjoy equal access. For example, one club good is
the electoral value of a party label. All members automatically enjoy (or suffer) a partys reputation or performance
in government.2
The second dimension of my typology is the extent
to which political parties control access to the benefits of
membership. In the case of private goods benefits, a party
might make all decisions about distribution. For example,
in most closed-list systems, ballot positions for upcoming elections may be considered a private good controlled
2
Club goods may not be entirely nonrival, however: each additional
member may slightly degrade the quality of the club good. For
example, as a partys size grows, it may become a catch-all party,
and its ideological label may lose meaning. New members with very
similar ideological positions will have minimal degrading effects;
legislators with distinctly different or even contradictory positions
may reduce the value of the party label dramatically.

by the party leaders. In other systems, private goods need

not be controlled by parties. For example, nominations in
the United States are still effectively private goods but are
controlled by voters in primary elections, not party leaders. With club goods, parties cannot restrict members
enjoyment of benefits. Again, all members enjoy a partys
brand name or reputation. Parties do control enjoyment
of club goods, however, when they can restrict membership to a subset of legislators, repelling or expelling unwelcome members. The framework proposed by Aldrich and
Bianco (1992) is a special case of type II, where politicians
compete in a single-member district for a private good (a
nomination or seat) without any interferance from party
leaders.
Legislators switch party to maximize their expected
career utility, a function of the resources they receive
from their party of choice, less a switching transaction
cost. Transaction costs can take many forms, codified in
law or imposed by informal institutions. For example,
Nepal, Ecuador, and Japan have passed laws whereby party
switchers are expelled from office, losing their mandates.3
Alternatively, transaction costs can be imposed by voters.
Where the electorate or party militants are very partisan
or otherwise use parties as information cues, switchers
will usually have little credibility and difficulty attracting
votes or campaign support.
Parties will offer resources and invite switchers when
the benefits of welcoming a new member exceed the costs
of resources offered. The value of a new member will vary
greatly across systems, depending on political institutions
and voters characteristics, but at least two benefits to accepting new members are obvious. First, in all legislatures,
size matters. On average, larger parties have more political
influence, more cabinet positions, better committee assignments, and more pork. The relationship between size
and resources may occasionally be linear, when resources
are distributed proportionately. In such a case, each additional member brings equal value in party resources.
But in winner-take-all systems and multiparty coalition
governments, the value of an additional member depends
on whether he/she changes the balance of power. Second,
switchers may bring with them electoral support for their
new party. Especially popular politicians, often celebrities, may significantly increase a partys vote share. Such
a benefit, however, depends on the electoral context. Voting must be personalistic enough that voters will continue
to vote for a switcher and the electoral system must pool
3
More recently, party leaders in Ecuador agreed to allow illegal party
switching to continue so a minority President could increase his legislative coalition size. The rule in Japan applies to legislators elected
through proportional representation and only prohibits switching
to a party that competed in their same district.

65

PARTIES FOR RENT

votes, so that other party members will directly benefit

from the influx of popular support.
Formalizing the Model. When formalized, this logic
makes predictions about the patterns and frequency of
switching across systems. I illustrate with a simple model,
following Aldrich and Bianco (1992). Let a legislature have
two parties, A and B, and two legislators 1 and 2. Legislator
1 starts in party A; legislator 2 starts in party B. Each legislator has an additional exogenous characteristic, g i , that
is their contribution or value-added to a club good of the
party they join. For example, a very popular politicians
might bring votes to a new party. g i need not be positive;
a scandal-ridden legislator might have a negative valueadded in a party.
Parties have exogenous resource endowments
j and
endogenous resource endowments f (G j ).
j may be a
private or club good. f (G j ) is a club good enjoyed by all
members. Its value is determined by f , a strictly increasing function and G

j , the sum of the g i of all members
g i j . When possible, parties exercise
of party j : G j =
control over access to resources by restricting membership. Specifically, members of each party cast simultaneous votes to determine which legislators will be allowed to join or stay and which will be refused admission. Party membersthemselves legislators in my
modelcast these votes to maximize their own expected
payoffs.
Legislators switch party or stay put to maximize their
expected utility, a function of the resources they will receive in their party of choice, less any transaction costs.
Legislator is utility associated with membership in party
j is:
ui j = i j
j + f (G j ) Ti I j =home
where
j is party js resources, i j is legislator is share of
party js resources, f (G j ) is the value of the public good
contribution of all party members to party j, and T i is
a transaction cost paid if legislator i switched party. The
precise form that payoffs and f (G j ) takes will vary with
the specific goods that contribute to parties value for and

contributions to legislators, and legislators value-added

for parties.
Two points deserve mention. First, the model is not
explicitly spatial, but can easily capture spatial or nonspatial payoffs. For example, letting  j be party js ideal
point, and  i be legislator is ideal point, let i

j = | j
 i | . Similarly, one could set f (G j ) = n j |
j g i |
where nj is the number of members in party j. In this case,
legislators payoffs are a function of the value of a party
label
j times their distance from the party. Similarly,
parties weigh the added benefit of an additional member
(nj + 1) versus the cost of increased ideological dispersion
|
j g i | when deciding whether to admit legislator i or
not. Second, in this version of the game, parties can only
invite/retain or reject/expel members. If payoffs
j were
nondiscrete private goods, the model could be relaxed so
that parties set i j legislator is promised share of party
js goods. Effectively parties could make continuous offers
to current and prospective members.
The game is played as follows:
1. If parties have control over membership, they simultaneously decide whether to accept or reject each
legislator i via majority vote of current members.
2. Legislators make membership decisionsswitching
or staying putand receive payoffs.
3. The game ends.
For the simple case with one legislator per party, the
payoffs of membership for each legislator are summarized
in Table 2. Legislators start in the upper-right hand cell:
(A, B). Note that where parties have control over access,
a decision to reject a member eliminates that legislators
options. For example, if Party A, controlled by legislator
1, decides to exclude legislator 2, the first column from
Table 2 is eliminated, and the game is reduced to legislator
1 choosing between Party A and Party B.
Virtually any standard game can emerge from combinations of parameter values, functional forms, and levels
of party control. When legislators can freely join any party,
i.e., parties cannot restrict access, the most common results include:

TABLE 2 Payoffs of Party Membership

Legislator 2

Legislator 1
Party A
Party B

Party A

Party B

1A
A + f (g 1 , g 2 ), 2A
A + f (g 1 , g 2 ) T

A + f (g 1 ),
B + f (g 2 )

B T + f (g 1 ),
A T + f (g 2 )

1B
B + f (g 1 , g 2 ) T , 2B
B + f (g 1 , g 2 )

66

SCOTT W. DESPOSATO

FIGURE 1 Party Membership Equilibria Strategies - Public Goods

without Party Control

Nobody Move: When T is big, both legislators stay put

in their current party.
All Aboard!: when the payoff differential |
A
B |
is big enough, both legislators join the resource-rich
party.
Battle of the Sexes: When f (g 1 + g 2 ) is big enough and
all g i positive, both legislators want to be in the same
party, but each prefers that other switch and pay the
transaction cost T i . Similarly, when
is a private good
or both g i are less than zero, i.e., each legislator has a
negative value-added, legislators may play BOTS but
prefer to be in opposite parties.
Fashion: One legislator prefers to share a party with
the other, but the other prefers to be alone. This result
emerges when one legislator has a sufficiently large positive g i and the other a sufficiently negative one, and
leads to a mixed-strategy equilibrium.
Get out!: Without party control, a legislator with a negative g i , starting in a resource-poor party could join
the other party and force the other legislator to aban-

don ship. The first legislator could then keep all the
resources of the wealthy party for herself.
In the case where parties can reject new members, the
range of parameters that lead to switching equilibria is
much smaller, and some are eliminated entirely. For example, for BOTS g i , Fashion, and Get Out!, the resourcerich party will simply reject unwanted new members.4
Figures 14 show predicted equilibria for different
types of goods
, different degrees of party control, and
different values of T and g i , with  = 1 for club goods
4

The market for party membership has clear parallels with labor
markets, with three important differences. First, endogenous parties, not firms, make personnel decisions. Second, personnel decisions are made to maximize members utility, not profits. Third,
instead of wages, legislators receive payoffs that may be private or
club goods. At this point, the model begins to bear a striking resemblance to the market for political science faculty: personnel decisions are made by majority vote to maximize departmental quality
and payoffs are partly private goods (wages) and partly-club goods
(reputation).

67

PARTIES FOR RENT

FIGURE 2 Party Membership Equilibria Strategies - Private

Goods without Party Control

and  = n1j for private goods, where nj is the number of

members of party j. Each of the four figures corresponds
to one of the ideal system types of Table 1. Within each
figure, the four plots show equilibria for different values
of T and G, as labeled. Combinations of
A and
B which
lead to All Aboard! are black, those
that lead to Nobody
Move are white, and
that lead to mixed or other switching equilibria are gray. Thus systems with switching have
large shaded regions; systems where switching is rare have
large blank regions. Code to produce similar graphs and
calculate payoffs for different types of goods and values
of all parameters is available on my Website.
Comparing the figures generates predictions about
the conditions under which we will observe party switching and the direction of such switching. The most important of these predictions include:
Increasing transaction costs eliminates switching.
Within all of the figures, moving from T = 0 to
T = 2 significantly reduces switching. For example,
in Figure 1, compare plot I with II. In I, there is always

unconstrained switching. In the plot immediately below that one. the range of
that lead to switching falls
to 64%. A similar vertical comparison of plots in almost
all figures yields different percentage changes, but identical resultsthe range of
that leads to switching falls
in every case.
Increasing resource differentials |
A
B | leads to All
Aboard! Both legislators join the well-endowed party, with
important qualifications.
Without party control, increasing party resource differentials leads to switching as legislators play All Aboard
or other strategies, for both kinds of goods. The range
of
that lead to switching is much larger for club goods
than for private goods, because club goods are nonrival.
With party control and club goods, the impact of resource differentials is unchanged. An important exception arises with party control and private goods. In this
scenario, all switching is less likely as party resource
differentials increase. When parties control valuable
private goods, they will not share them unless a new
member brings a g i that exceeds the cost of sharing.

68

SCOTT W. DESPOSATO

FIGURE 3 Party Membership Equilibria Strategies - Public Goods

with Party Control

In the simple illustration, parties only accept switcher i

when g i > 2j .
Negative g i decrease switching; positive g i increase
switching
Where parties control access and g i is positive, party j
will accept new member i, except when
j is a private


good and 2j > g i . But where g i is negative, parties will
always exclude that member if they can.
Where parties cannot restrict membership, a negative
g i reduces the range of All Aboard! and increases the
ranges of mixed-strategy and other equilibria discussed
above.

consider the institutional and societal incentives that create a very fluid market for party membership. Key features
of this market are relatively low transaction costs, payoffs
that are either club goods or uncontrolled private goods,
large resource differentials, and positive value-added for
most legislators. Effectively, most Brazilian deputies fall
somewhere in Figures 1 and 3, Plots I and II. The parameters of the system create and structure many opportunities
for defection from parties.

Adding more legislators and more parties increases the

complexity of the model considerably, but the basic implications remain unchanged.

Brazil has some of the highest rates of party switching in

the worldexceeding one-third of deputies during the
average legislative session. In some countries, switching
appears to be a short-term transitional phenomenon associated with democratization and the gelling of a new
party system or a short-term political realignment. This is
not the case in Brazil: switching is apparently chronic and
enduring. Table 3 shows switching rates in the last several
legislatures. During the 49th legislature (199194), there

An Application to Brazil
In this section, I apply the model to the case of Brazil. I
begin with an overview of switching in that country. I then

Switching in Brazil

69

PARTIES FOR RENT

FIGURE 4 Party Membership Equilibria Strategies - Private

Goods with Party Control

TABLE 3 Party Switching Rates in the Chamber

of Deputies
Legislative Session

Number of Switches
Switching Rate
Number of Switchers

49th

50th

51st

52nd

262
.52
198

212
.41
169

262
.51
183

211
.41
153

Projected rates for current legislature, as of October 2005.

were 262 incidents of switching, for an average switching

rate of .52 per legislator-term. That figures slips slightly to
.41 in the second period, rises to .51 in the third, and rises
again to zzz in the current period. Further, while switching
was quelled by Brazils military regime, it was apparently
common during the Second Republic (194664) and there
is even evidence of frequent switching during the First
Republic (18891930; Graham 1990)

Switching has not escaped the attention of students

of Brazilian politics, though previous work has focused
primarily on the impact of switchingtreating party
membership decisions as an explanatory variable. For example, Nicolau (1996) examines the impact of switching on party size, finding that switching changes relative
party and coalition strength in the Chamber of Deputies.
He speculates that switching is prompted by ideological
conflict, personal disagreements, or electoral opportunities, but suggests that separating these empirically would
be impossible. Mainwaring and Lina n (1997) examine
patterns of switching and party discipline in the Constituent Assembly, finding that switchers tend to have
lower than average agreement scores in their party of origin, and substantially higher agreement scores in their
destination parties. Schmitt (1999) examines the impact
of switching on reelection rates for legislators, reporting slight negative effects on electoral success among
switchers that seek reelection, but without controlling
for strategic candidacies. And Melo (2000) provides a

70

SCOTT W. DESPOSATO

descriptive look at party switching from 1985 to 1999,

finding that switching is more common among conservative than leftist parties, though switches from both sides
can traverse the political spectrum. He concludes by challenging the notion that Brazils parties can be considered
cohesive or institutionalized.
My goal is to explain party affiliation patterns and
shed light on the Brazilian party system. Applying my
theoretical framework to the case of Brazil requires considering the nature of payoffs, transaction costs, and legislators value-added in the Brazilian context. As I discuss
below, for many Brazilian legislators, transaction costs
are very low, value-added is usually but not always positive, and key payoffs of membership are not pure private
goods. Consequently, most parties are willing to accept
new members, and legislators are prone to switch when
resource differentials are large. There is variance on these
general themes across legislators and parties.

Transaction Costs
Transaction costs are manageably low for many Brazilian
legislators. The legal environment is very accomodating:
there are no legal prohibitions on party switching, except immediately prior to elections for those seeking reelection.5 Until recently, all incumbents were guaranteed
renomination in the next election under the candidato
nato or birthright rule.6 Further, personalistic voting
means that switchers can take their voters with them.
Partisanship in Brazil is relatively low (Samuels, forthcoming), and most switchers I interviewed expected no
or little repercussions in terms of electoral support. Many
seemed to think that voters simply didnt care about party
membership. In one interview, the legislators receptionist
wasnt even aware that he had switched parties!
Some legislators, however, face significant transaction costs for switching. A few Brazilian parties have aggressively cultivated partisan followings, building grassroots organizations, promoting cohesive policy platforms,
and encouraging partisan, not personal votes in campaigns. Most notable of these is the Workers Party, or PT,
but some smaller parties (including the Green Party) also
have more ideological popular followings (Ames 2001;
Keck 1995; Mainwaring 1999; Samuels, forthcoming) and
even catch-all parties may have a few committed partisan
supporters. Legislators with such constituencies cant take
5
Candidates for office must stay in a single party in the months
leading up to the election.
6

This provision was recently eliminated by a court decision but was

in effect for the period covered by my dataset.

their voters with them when changing party and should

be much more likely to stay put.

Benefits of Party Membership

Brazilian parties can facilitate access to pork, provide ideological or policy benefits, and can directly facilitate electoral success. One of the unique features of the Brazilian
system is that the primary benefits of party membership
are either club goods or uncontrolled private goods, facilitating a fluid party system. Ill discuss the nature of
benefits and parties role in their distribution in the following paragraphs.
Pork. A major benefit of party membership is priveleged
access to state resources. In the case of Brazil, scholars have
widely acknowledged the importance of state resource dis
tribution for deputies careers (Abrucio
1998; Ames 2001,
1995a, 1995b; Leal 1977; Mainwaring 1997, 1999; Pereira
and Mueller 2004; Samuels 2003). Government resources
can be channeled into pork projects that provide lucrative
construction contracts and needed infrastructure. Cushy
jobs in the bureaucracy reward campaign workers and
maintain politicians support networks for the next election. Resources might also be used in less legal waysfor
direct campaign activities or even vote buying.
Public resource distribution is controlled by the executive branch, not political parties. For legislators, access
to public funds and resources is maximized by joining the
governing coalition and voting for the executives proposals. Brazilian Presidents typically use state resource distribution to build multiparty supermajority coalitions. Supermajorities are required because legislators, even in the
governing parties, are hesitant to support controversial
economic reforms that impose significant costs on their
constituents.7 Finally, there are stark differentials in access to resources across parties. Legislators in opposition
parties are likely to receive significantly less pork for their
constituents.
Two empirical implications follow. First, government
parties should retain and attract members, given large
resource differentials and open market for votes and
pork. Second, since pork is a rival good, governing parties attractiveness may decline as the coalition reaches a
saturation point.
A similar logic applies to subnational coalitions.
Brazils decentralized federalism grants subnational political actorsespecially state governorscontrol over
7

For the executive, supermajorities may well be cheaper, allowing

the president to pay lower costs on a vote-by-vote basis to buy a
majority.

PARTIES FOR RENT

resources that national legislators covet (Abrucio

1998;
Mainwaring 1999; Samuels 2003). State governors influence should only be enhanced by progressive ambition
in Brazil. Many national deputies aspire to state and local elected and appointed positions. Governors directly
control key state appointments and can facilitate election
to subnational office by providing campaign resources,
endorsements, and selective pork distribution. Consequently, governismoa tendency to join the governing
coalition to enhance resource accessought to extend to
subnational governments, making state governors parties more relatively more attractive than state opposition
parties.
Electoral Payoffs. Brazilian parties play a limited, but important role in elections. In many political systems, parties provide many electoral resources: campaign workers,
financing, consulting services, and meaningful labels to
cue voters. But most Brazilian parties play little or no active role in legislative campaignsparties do not provide
candidates with volunteers, campaign finance, or consulting servicescandidates must generate these resources
on their own. Party labels are generally not used by most
voters as information cues; most citizens cast votes based
on candidates reputations, not based on their partisanship (Mainwaring 1999, 122). But party membership still
can have an important effect on electoral outcomes. A
quirk in the Brazilian electoral system, open-list proportional representation (OLPR), makes election easier in
some parties than in others and affects legislators affiliation incentives.8 The OLPR vote-pooling mechanisms
allow the minimum number of votes required for election
to vary across party and state. For example, a single, very
popular candidate could draw millions of votesearning
several seats for her party. Because of vote pooling, her extra votes could make election easier for other candidates
within her party.
Consequently, legislators have incentives to switch
party to take advantage of easier electoral quotas. Parties
where election is easier should retain members; parties
where election requires more votes should lose members.
This should vary with legislators own electoral strength.
Marginal legislators should be concerned about their
partys electoral quotient. Politicians that barely passed
the electoral threshold should seek another party where
8

Under OLPR rules, Citizens cast a single vote, usually for an individual candidate. Seats are distributed to parties based on the total
of votes received by all of the parties candidates. Mandates are
distributed to candidates based on the total number of votes each
receives. See Ames (2001, 1995a, 1995b), and Mainwaring (1999)
for more details on OLPR.

71
they could be elected with a safer margin. At the other
extreme, strong vote getters dont need to pay any attention to quotients; their election would be assured in any
party. In interviews, legislators have described these calculations: comparing their own expected vote share with the
electoral quotient in each party and switching to enhance
reelection prospects.
Electoral quotas are effectively uncontrolled club
goods. Enjoying a partys quota just requires joining the
party. Ballots are unordered, so legislators do not compete for list position. And every party has a surplus of
nominations: parties can nominate more candidates than
there are seats in every district, and rarely find enough
willing candidates. Consequently, legislators should adopt
membership strategies to take advantage of party quota
differentials.

Ideological Payoffs. A third component of party affiliation payoffs is ideological. Membership in a party of
like-minded legislators has several payoffs. Political parties can play an important role in helping like-minded
legislators advance a policy agenda. In addition, a political party acts as a brand name, providing information to
voters about a legislators policy positions (Aldrich 1995;
Snyder and Ting 2001). Further, membership in an ideologically incompatible party can have significant costs.
Legislators will be less able to advance their preferred
agenda, potentially angering voters and frustrating legislators own policy goals. Parties might also punish their
less-compatible membersgiving them fewer advancement opportunities, refusing to support their proposals,
or even preventing them from speaking on the floor. As
a result, legislators should prefer membership in ideological proximate parties and avoid those far from their
own ideal points. Ideological compatibility is clearly a club
good. All legislators enjoy or suffer their compatibility
with their copartisans. Differentials in ideological agreement should prompt legislators switching to, or staying
in, the most compatible party.

Pork or Policy ?. Legislators may have to balance competing desires for state resources and ideological compatibility. In such situations, their behavior should reflect
constituents demands. Legislators from less-developed
regions should prioritize access to state resources to deliver the public infrastructure their constituents need.
Legislators from more-developed regions should be
more responsive to ideological concerns, as basic infrastructural needs have been satisfied. This logic implies

72

SCOTT W. DESPOSATO

interacting pork and ideology differentials with an index

of development.

maximizes their expected payoff, a function of party resources less transaction costs. In a multiparty environment, we can write this as:
Uijt = MAX(Ui 1t , Ui 2t , Ui 3t , . . . , Uimt ).

Legislators Value-Added: g i
With important exceptions, most deputies have a positive impact on f (G) for most parties. In the Chamber
of Deputies, size matters. Most resources are allocated
proportionally: committee assignments and chairships as
well as Chamber leadership positions are a direct function
of party size. In addition, for members of the governing
coalition, commanding additional votes usually translates
into additional cabinet positions and government jobs
when negotiating with the executive branch. The important exceptions include the several Brazilian parties that
rely more on their ideological cohesion and policy label
to generate votes than their capacity to generate pork. For
such parties, ideologically inconsistent legislators have a
negative g i and should be denied admission.

Summary of Predictions
The preceeding discussion offers the following hypotheses
regarding party membership in Brazil:
H1 : Increasing transaction costs decreases switching.
Legislators with more partisan voters will be less likely
to switch party.
H2 : Increasing resource differentials increases switching into resource-rich parties.
State resources: Legislators stay in, or switch into governing parties, but the attractiveness of such parties may
decline as coalition size grows.
Electoral Opportunity: Legislators stay in, or switch into
parties with easy electoral thresholds.
Ideology: Legislators stay in, or switch into ideologically
compatible parties.
Pork or Policy: Legislators from less developed regions
are most responsive to resource access; legislators from
developed regions are more resposive to ideological
concerns.
H3 : Parties exclude legislators with negative valueadded.
Switchers will be welcomed when their value-added is
positive but excluded when ideologically too inconsistent with a party.

Methods
The formal model can be operationalized empirically as
a discrete-choice problem. Legislators pick the party that

where uijt is legislator is utility of membership in party p

at time t. When party j is her current party, she stays put.
When j is some other party, we observe a switch. I write
legislator is utility associated with membership in party j
as:
ui j t = 0 + 1 X i j t 2 Ti I j =home + 3g i + i j t .
where X ijt is a matrix of the covariates described above,
and 1 is a vector of coefficients. If we assume that the
 ipt are independent and identically distributed following
an extreme-value distribution, then this becomes a standard conditional logit model, also known as McFaddens
choice model. Though rarely used by political scientists,
the conditional logit has been widely used by economists
to explore individuals choices between products, transportation options, and, most comparable to this project,
migration patterns.9

Measurement
To test my hypotheses, I collected data on party membership, constituencies, and party characteristics for all
deputies during the 49th and 50th legislatures. Estimating coeficients requires measuring the benefits that each
legislator expects to receive from each party. The unit of
analysis is the legislator-month, that is, each legislators
membership choice during each month in office.10
9

See McFadden (1973), Maddala (1983), Long (1997), and Long

and Freese (2001). Alternative choice models can be implemented
with more flexible error distribution assumptions, but are only
tractable for smaller choice sets and datasets. See Glasgow (2001)
for more details.
An alternative, and possibly more accurate, specification would
incorporate parties valuation of legislators separate from deputies
choicesa model where parties first decide whether or not to accept
a legislator, then legislators choose from the choice set available to
them. I have a working paper demonstrating how to estimate such a
model, but it proved intractable for this dataset (Desposato, 2005).
10

Why is it necessary to use a time-based measure of affiliation

decisions? First, many of the key variables in the modelcabinet
formation, committee assignments and leadership, party leadership, and even the party of the Presidentsvary over time, as do
many of the control variables. During one period, legislator is party
might have cabinet access. A month later, after a cabinet reshuffle,
legislator i might be in the opposition. Second, the actual membership of the Chamber of Deputies is constantly in flux as legislators
leave and substitutes take their places. Using the legislator-month
as the unit of analysis accounts for the varying amount of time that
each legislator actually served in the Chamber.

PARTIES FOR RENT

73

Dependent Variable: Membership at Time t. The dependent variable follows standard practice for choice models.11 Y ijt is legislator is choice of party during month t,
coded 1 if the legislator chose party j and 0 otherwise.

GovDiff = Gov ij Gov ik

Govij is coded 1 if prospective party j was in the governors office in legislator is home state, and 0 otherwise.

Electoral Threshold. I measure electoral payoff differentials with Tdiff . I first calculated Threshold for each
legislator-party combination, coded 1 if legislator i
would have been elected in party j, and 0 otherwise.
Finally, I calculate the threshold differential as:

Ideology. I test the impact of ideological compatibility

differentials on party membership using DistDiff ij , the
relative proximity of prospective party j versus current
party k.12 Ideal points are estimated using WNOMINATE
method (Poole and Rosenthal 1997).

Tdiff = Threshold ij Threshold ik ,

where i references the legislator, j references the prospective party, and k references the current party.
This measure captures both the electoral value of a
party and the marginality of the legislator. Some legislators would have been elected in any partyso party
membership does not affect their electoral prospects directly. For such deputies, Tdiff = 0 for all parties. For
weaker legislators, party strength matterssome parties
reduce electoral prospects; others increase them, captured
in the negative or positive values of Tdiff .
State Resources. Measuring the pork and other resource
benefits of party membership in Brazil can be quite difficult. Because Brazil uses multimember district elections with high district magnitude, scholars cannot easily tie government expenditures to individual legislators.
One alternative would be to use budget amendments,
but amendments are frequently ignored by the executive
branch and are only a small part of government expenditures. In fact, the executive branch at all levels of government has very strong control over budget development,
expenditures, and other components of resource distribution (Schneider 2001). Most real access to financial resources is gained through individual legislators lobbying
of the President and Ministers, consistent with patterns
observed in other countries.
How can we measure access to government resources?
Ames (2001) creatively used meetings with ministers as
a proxy for pork access. As no such data is available for
my time period, I use two unbiased, if crude, proxies for
resources access: coalition membership.
CabDiff = Cabinet ijt Cabinet ikt .
Cabinetijt is coded 1 if party j was part of the presidents coalition at time t, and 0 otherwise. As above,
j references the prospective party and k references the
current party.
11

See Maddala (1983), Glasgow (2001), Long (1997), and Ben-Akiva

and Lerman (1985).

Pork or Policy ?. Which of these payoffs matters most

to legislators will depend on the underlying preferences
of their constituents. As a proxy for preferences, I use
Educationi , the percentage of legislator is constituents
with four or more years of education. I interacted this
variable with CabDiff, GovDiff , and DistDiff . Education
is highly correlated with other demographic indicators
of development. A well-known pattern in Brazilian politics, as in many other countries, is that pork, patronage,
and clientelism are more common in less-developed regioned, and policy promises are likely to be the currency
of politics in more developed regions. CabDiff and GovDiff should have their strongest impact in low education
regions; DistDiff should have its strongest impact in high
education regions.
Transaction Costs. Precisely identifying the partisanship
of each legislators vote share is impossible, given the lack
of detailed survey data and the high district magnitude
in Brazil. We can, however, measure the degree to which
each party has a partisan following using list votes. Voters
have two choices in Brazil: to cast votes for individual
candidates or to cast votes for the overall party list. The
great majority (over 90% in 1994) cast personal votes, but
this varies substantially across party. In the center-right
Liberal Front Party (PFL), only about 1% of the votes
were list votes, while the Workers Party (PT) received
almost 34% list votes. I use two measures of partisanship:
Municipal List Voting (M i ) and National List Voting (N i ).
For each, I take advantage of the structure of the Brazilian
ballot: voters can cast votes for individuals or for the party
list.
M i is the percentage of legislator is electoral base that
cast party-list votes. Specifically, I calculated an average of
list voting in the areas where each legislator received votes,
weighted by what proportion of their votes were earned
12
Specifically, Distance ip is calculated as DistDiff ij = |  i  j |
|  i  k | , where  i is legislator is ideal point,  j is the centroid
of prospective party j, and  k is the centroid of current party k.

74

SCOTT W. DESPOSATO

there. This measures partisanship among each legislators

subconstituency within a district. N i is the percentage
of list votes received by the legislators party, based on
national returns. Together, these proxy partisanship. M i
captures characteristics of the voters in is electoral base;
N i controls for the size of a partys following in aggregate.
Because both variables were highly skewed, I used their
natural logs in all models.

Legislators Value-Added: g i . All legislators make a small

positive contribution to all parties, but deputies that are
ideologically incompatible will have a negative impact on
the partys policy label. This implies that a deputy is valueadded in party j is:
|i  j |
gi = c
j
where c is the constant value-added that all parties receive,
 i is legislator is ideal point, and  j is party js centroid,
and  j is the standard deviation of the ideal points of
party js members. Thus, being far from a diverse catchall party (large  j ) is not problematicthe second term
is small and g i is positive. But being far from a highly
cohesive and unified party (small  j ) is problematic
g i is negative. Unfortunately, this cannot be estimated
separately from payoffs. The first term, c, is a constant.
The second term is too highly correlated with ideological
compatibility to be estimated separately. Effectively, this
means that the DistDiff term must do double duty in
the modelcapturing both legislators payoffs and their
value-added.
Controls. Building on existing work on party switching,
Brazilian politics and quantitative methods, I included
several additional variables as controls:
PRN controls for the anti-PRN stigma associated with
that party, as President Collor was impeached for corruption in 1992. PRN is a dummy variable coded 1
for that party from the time the scandal broke until the
end of the legislature (August 1991December 1994),
and 0 otherwise.
Home is a dummy variable identifying legislator is current party, effectively a constant controlling for the
baseline frequency of switching.
Comit is coded 1 if legislator i was on a key committee,
or a committee leader, at time t. Com should have a negative impact on switching: committee assignments are
controlled by party leaders and may be lost by switching.

PartyLeader it : Coded 1 if legislator i was a leader or a

vice leader of her party within the legislature at time t,
and 0 if she was not. Party leadership provides prestige, additional staff and office resources, and influence
over the legislative agenda. Deputies in leadership positions should be less likely to switch.
Incumbent is a dummy variable coded 1 for incumbents and 0 for first-term legislators. More experienced deputies should be less likely to change party.
I also included several time-specific variables to control
for key moments and differences in the legislative cycle
and to allow the underlying hazard rate of switching
to vary during the legislative session (Beck, Katz, and
Tucker 1999).

Results
Tables 4, 5, and 6 show coefficient estimates and predicted
probabilities for models of party affiliation. The results

TABLE 4 Conditional Logit Model of Affiliation

Decisions
Model 1

Payoffs
Threshold
Cabinet
Cabinet Educ.
Governor
Gov. Educ
Ideol Distance
Ideol Dist. Educ
Transaction Costs
District Partisanship
National Partisanship
Controls
Committee
Incumbent
Party Leader
National Campaign
Election Deadline
Orgnize Period
50th Leg.
PRN
Home
N
LL

Model 2

Coef.

SE

Coef.

SE

.88
1.14
.80
1.51
1.86
.79
.15

.32
.28
.53
.35
.69
.37
.64

.89
.73

.32
.09

1.62
2.09
.71

.34
.67
.13

.40
.49

.07
.07

.40
.49

.07
.07

.02
.19
.45
.89
1.88
.91
.83
1.30
12.51

.12
.13
.16
.25
.19
.27
.19
.31
.51

.46
.88
1.89
.91
.81
1.33
12.62

.16
.25
.19
.27
.18
.31
.51

43,286
1981.47

Cubic spline estimates not shown.

43,286
1983.76

.05 .01

75

PARTIES FOR RENT

TABLE 5 Impact of Selected Party

Characteristics on Predicted
Probability of Membership

TABLE 6 Impact of Transaction Costs on

Probability of Switching
P(Switch)

Electoral Threshold
National
Below

Above

0.29
(0.18, 0.42)

0.71
(0.58, 0.82)

Cabinet Membership
Opposition

Cabinet

0.32
(0.29, 0.37)

0.68
(0.63, 0.71)

Ideological Distance
Far

Close

.14
(0.07, 0.23)

.86
(0.77, 0.93)

State Coalition
Education

Opposition

Governor

Low

0.21
(0.14, 0.31)
0.53
(0.40, 0.66)

0.79
(0.70, 0.86)
0.47
(0.34, 0.60)

High

95% confidence intervals in parentheses.

show that my affiliation model fits the Brazilian case quite

well. The supposedly chaotic and antiparty behavior of
legislators in actuality reflects calculated and rational career moves. In particular, the data suggest that Brazilian legislators are seeking distributive, ideological, and
some electoral opportunities, but institutional resources
are not highly valued. Further, party-switching patterns
vary with constituency types. Characteristics of the electorate raise and lower transaction costs, and shift legislators priorities between ideology and distributive functions
of parties. I discuss the specific results in the following
paragraphs.
First, there is evidence that legislators use Brazilian
parties as electoral vehicles. Brazils electoral rules can
make election easier in some parties than in others, and
party affiliation patterns take these rules into account.
The coefficient on Threshold is positive and significant:
legislators choose parties where vote pooling will facilitate
reelection, and avoid parties with thresholds that are too

District
Low
High

Low

High

0.061
0.013

0.010
0.002

high. Table 5 shows the substantive impact: legislators

more than twice as likely (p = .71) to choose a party
with an easy electoral threshold than one where additional
votes are needed for reelection (p = .29).
Second, legislators value ideological compatibility
when choosing between parties. Deputies dont want to
join or stay in a party that is far from their own policy
preferencesthe coefficient on Ideol Distance is negative
and significant. The impact is substantial: deputies are
about five times more likely to pick an ideologically proximate party than one at the opposite end of the political
spectrum (p = .86 versus p = .14). In other words, the
farther a deputy is from a party, the more likely she is to
switch away if she is a member, and the less likely she is to
switch to that party if she is not. This finding is evidence
that legislators do care about ideological consistency and
compatibility.
These results would hardly be surprising in most
legislatures; they are potentially controversial for Brazilianists. Brazils largest parties have been categorized as
inchoate and catch-all: why should voting coalitions matter for legislators in such parties? The importance of roll-call compatibility may reflect legislators
own desire to be ideologically consistent, pressure from
party leaders, or pressure from constituents. Which is
at work cannot be parsed out here, but the key point
is that Brazilian legislators do value parties as voting
coalitions.
Third, the analysis confirms the importance of distributive capacity for legislators careers; the coefficient
for Cabinet is positive and significant. Legislators in parties with access to federal government largess are more
likely to stay put, and switchers are more likely to head
for such parties. When choosing between two otherwise
identical parties, legislators are about twice as likely to
pick a cabinet party than an opposition party: pCabinet =
.68 versus pOpp = .32. This is consistent with existing
work on the importance of patronage in Brazilian politics (Amorim Neto 1998).

76

SCOTT W. DESPOSATO

Governors Party is positive and significant, and the

interaction of Education and Governor is negative and significant, as predicted. In impoverished and less-developed
areas, voters prioritize governismofollowing the state
governor in exchange for access to state resources. In such
areas, legislators in a gubernatorial party are more likely to
stay put, and switchers more likely to make such a party
their destination. This finding shows that decentralized
Brazils state politics spill over into the national arena,
affecting party affiliation decisions and coalition formation. These results echo previous research on the impact

of federalism on national politics in Brazil (Abrucio

1998;
Mainwaring 1999; Samuels 2003), with an important additional point.
Gubernatorial influence varies with constituency
type. A deputy with a low education constituency is about
three times more likely to pick a state governors party
over a state opposition party (p = .79 versus p = .21).
But among high education constituencies, legislators are
effectively ambivalent between state opposition and governing parties: p = .53 versus p = .47, respectively.13 In
impoverished and less-developed areas, where voters prioritize governismo, legislators align with the state governor
in exchange for access to state resources. In more developed areas, voters demands for localistic goods decline.
This decline reduces the relative attractiveness of stateruling parties and increases the relative importance of
ideological and other factors in affiliation decisions.
The other interaction effects were not significant. The
Education Cabinet interaction is negative, as predicted,
but its effect is not significant. It may be that the benefits of membership in executive coalitions vary across
governmental levels. Perhaps cabinet membership provides payoffs to legislators with all kinds of constituencies, while gubernatorial affiliations are most important
in backward, clientelistic contexts. The other possibility
is collinearity between Governor and Cabinet status
parties in the Cabinet also tend to be parties winning gubernatorial elections. I also predicted that the interaction
of Distance and Education would be negativereflecting
an increased attention to ideological consistency as voters preferences change. However, the interaction effect
is small, has the wrong sign, and is not significantly different from zero. Two interpretations are possible. The
first is that my argument about voters does not apply to
voting coalitionsperhaps all legislators are equally concerned with ideological consistency, regardless of their
constituents preferences. The other interpretation is that
legislators with more educated constituents have already

self-selected into the parties that are ideologically most

proximate.
Finally, the results emphasize the importance of voters in creating strong and stable party systems. Legislators
with more partisan constituencies are much less likely to
ever switch party. Table 6 reports the probability that a legislator switches party during a four year legislative term
as a function of partisanship. In all cases, the choice set
is restricted to two identical parties, so these probabilities
can be interpreted as the impact of random, unobserved,
within-party personality conflicts and struggles that cause
switches. The hypothetical party list voting levels are set
to 0.6% (low) and 30% (high) which is slightly narrower
than the actual range of the data. Legislators in less partisan parties, with personalistic constituencies, have a .061
probability of switching during their term, even when
there are no payoff differentials. At the other extreme,
legislators with more partisan electoral bases are very unlikely to switch, with a predicted probability of switching
of just .002 over a four-year period.
The control variables are mostly as expected. Incumbency may deter switchingits coefficient is negative
though it does not reach conventional levels of significance. Party leaders, not surprisingly, were significantly
more likely to stay put than nonleaders.
The results also have important implications for parties internal organization of the Brazilian legislature.
Scholars have argued that party leaders control of the
mechanisms of the Chamber of Deputies enables them
to enforce discipline and maintain party cohesion. Of
all resources in the Brazilian legislature, party leaders
have most direct control over committee assignments.
Apparently, however, this control is nearly worthless.
Good committee assignments do not deter switching:
the coefficient on Com is not distinguishable from
zero.
The time-specific variables did shape switching frequency. Legislators were more likely to switch party
around campaign filing deadlines, and significantly more
likely to stay put during campaigns. The coefficient on
Organize and Deadline are positive and significant, indicating that switching is more frequent during the first part
of each session when coalitions, committees, and leadership are forming, and immediately prior to new election
filing deadlines.

13
All discussions of interaction effects significance account for the
covariance between the base and interactive terms.

Frequent party switching is usually treated as a simple

indicator of system weakness, but can also provide a

Discussion

77

PARTIES FOR RENT

powerful and unique window on legislative behavior, allowing us to understand the functions that parties serve
for ambitious politicians. I built a model of party affiliations where legislators choose parties to maximize electoral, institutional, ideological, and distributive opportunities, less a transaction cost for switching. The model
and method could be applied to any system, and when
adapted to Brazil in particular, several findings emerge.
First, legislators in Brazil use parties for electoral,
ideological, and distributive ends, but not for institutional advancement. Legislators prefer membership in
parties that maximize their chances of reelection, that
have more access to pork, but that are also ideologically compatible with their own goals. Ironically, legislators affiliation choices were not affected by committee
assignmentsthe one resource that Brazilian party leaders firmly control. At the same time, stabilizing pressure
comes from an unexpected source: voters. While previous work has downplayed Brazilian voters influence on
legislators, I found that deputies with more partisan constituencies face significant defection-deterring transaction costs. Legislators with partisan constituencies were
substantially less likely to switch party than those from
the least partisan regions and labels.
Further, which factor matters most varies with voters preferences. Higher education voters shift legislators
efforts toward ideological concerns; lower education constituencies have other priorities, and their representatives
are more likely to emphasize access to government largess
in their party-affiliation strategies. Scholars have found
that parties play various roles in legislative systems, creating policy labels, organizing governments, mobilizing
voters, and providing campaign assistance. The results
from my research show how the roles parties play for
politicians are not static within a political system but can
vary with electorate preferences. More simply, legislators
with different kinds of constituents use parties to very
different ends.
Perhaps most importantly, the results show how representation works in spite of party switching in Brazil.
Party switching may be viewed as a challenge to representation when voters use party labels to cast ballots and pick
policy platforms. Switching effectively destroys the meaning of party labels, raises voters information costs, and
eliminates party accountability. Switching can be viewed
as a threat to the very core of democratic representation.
But ultimately, it appears that switching is not the threat
it appears to be, at least in the case of Brazil. Where partisanship is high and voters rely on party labels to pick
candidates, transaction costs make switching very rare.
Further, when legislators do switch, they are likely to move
to a party with a similar ideological positionpreserving
the policy value of a vote. Finally, patterns of affiliation

change to match voters preferences for different kinds

of goods. Where voters prioritize local and particularistic
goods, legislators respond by switching to the parties that
will maximize their ability to deliver.
My approach should be applicable in other countries.
The obvious task is to carefully identify the benefits of
party membership that could affect legislators career utility, taking into account each systems institutions, norms,
and transaction costs. Other political systems may offer
very different incentives for the frequency and patterns of
party switching, but based on my results, electoral rules,
patterns of legislative organization, and voter partisanship and preferences are likely suspects for examination.
Just as important, however, is a careful consideration of
the nature and scarcity of payoffs of party membership.
Am important distinction is the role that parties play
in controlling membership and distributing goods. In the
case of Brazil, the primary payoffs to membership are either club goods or private goods that parties do not control. But where parties do exercise control over resources,
scholars will need to account for the interaction between
parties and legislators.
These distinctions have important implications for
the fluidity of the market for party membership. When
party resources are scarce and controlled by parties,
switchers need to prenegotiate access to them. For example, the U.S. House of Representatives winner-take-all
distribution of committee chairships makes those positions scarce private goods. Consequently, representatives
have to prenegotiate their seniority before switching, as
did Billy Tauzin (DR/LA) before he changed to the Republican Party in 1995, and Ralph Hall (DR/TX) in
2004. Hall retained his seniority and committee assignments, and Tauzin actually became the Commerce Committee Chair after Thomas Bliley (R/VA) stepped down.
In addition, incorporating strategic party behavior also implies broadening our focus from legislators
staying put or switching party to also including party
expulsions as part of legislators and parties utilitymaximizing behavior. Expulsions, like party-switches, are
relatively rare but not unknown. Legislators have been expelled for corruption and inappropriate use of party resources (Mexico and India), failure to close ranks and endorse party candidates and positions (United Kingdom),
not being a real communist (Ukraine), helping party
dissidents (Ethiopia), and supporting an unpopular policy (conscription, Australia). There are many other examples.14 Including expulsions will provide a more complete picture of the market for parties and legislators
14

Indeed, there have even been a handful of recent expulsions in

Brazil, nearly all from the ideological leftist parties. There were too
few during the period I studied to incorporate them here.

78

SCOTT W. DESPOSATO

and also enrich our understanding of parties collective

preferences.15
Future work will face two methodological challenges.
First, incorporating parties as strategic gatekeepers, and
adding expulsion to their arsenal will require more complex methods than those used in previous work and in
this article. Specifically, empirical work in such systems
should explicitly model the decision to accept or reject
new members as a first-level selection decision that determines the choice set available to legislators at their decision stage (Desposato, 2005). Second, in other cases there
may be less information available about the payoffs for
switching. If a legislator chooses party 1 over party 2, we
know that U i 1 >U i 2 . But while we may observe legislator
is subsequent benefits of membership in party 1 (renomination, a good list position, ideological compatibility),
in some cases we do not have much information about
legislator is expected list position or renomination status
in a party not chosen. Again, in the case of Brazil, the primary benefits of membership are easy to estimate, even
for parties not chosen. But in some cases, negotiations
on switches that did not happen may be private information. A Bayesian approach, treating the unobserved
components of membership benefit as missing data, or a
selection model would solve this information problem.
Regardless, this article shows how party switching
provides a valuable window on the nature and functioning
of party systems. Switching is frequently treated as simply an indicator of system weakness or failure. Ive shown
here that switching is a systematic phenomenon, best understood as the rational behavior of career-seeking politicians in the absence of stabilizing transaction costs. In
Brazil, switching has led scholars to describe that country
as dominated by parties for rent. My results show that
while parties may be for rent, the market for membership
is highly structured, rational, and constrained by both
voters and institutions. The approach I advocate is easily generalizable and could yield insights for many other
systems.

Appendix
Additional Information on Variables
This appendix provides additional information on the
coding of variables.
15
Labor economists have argued that there is effectively no difference between voluntarily quitting a job and being fired or laid off
(McLaughlin 1991). Firms would be happy to continue employing
laid-off workers if said workers would accept drastic pay cuts.

The dependent variable is party-membership choice

of legislator i in month t. In a very few cases, legislators
switched out of one party but remained without party
membership. If they quickly moved to a new party, I ignored their time in no-partys land. Only a handful of
deputies failed to ever join a new party, so I excluded
them from the analysis.
If legislators changed party more than once during a
month, I only counted the first switch. Again, almost no
deputies made more than one switch per month. Also, a
legislator in theory could start a new party or join a party
with no representation, but practically all deputies moved
into an existing party.
There were several party mergers and splits between
1991 and 1998. During the 49th legislature, the PCB became the PPS, the PDC and PDS became the PPR, and the
PST and PTR merged to become the PTR. In the second
period, the PPR, PRP, and PP merged to form the PPB,
and the PCB changed its name to the PPS. I did not count
any of these as party switches. For example, if a deputy
from the PDS transferred to the PPR with the rest of her
colleagues, I coded this as a stay-put.
I measured constituents education level using data
from the 1990 Brazilian Census, specifically, the percentage of adults with more than four years of education.
Legislators constituents education levels were calculated
by taking the average education level from each municipality where the legislator received votes, weighted by the
percentage of her votes received there:
n

MuniEd m Votes im .
AvgEd i =
m=1

Education data came from the 1990 Census. To account for the creation of some 500 new municipalities in
1993, I reaggregated vote shares from the 1994 election
back to the 1990 municipal boundaries, then used these
aggregated vote shares to calculate voter-education levels.
I used a similar approach to calculating T 1 , a measure
of the partisanship of legislators constituents. I calculated
the percent party-list voting in each municipality, then calculated an average of these municipal-level partisanship
measures, weighting by the percentage of each legislators
votes received there:
n

MuniList m Votesi m .
Mi =
m=1

Legislators ideal points were measured using the

Poole-Rosenthal NOMINATE method (Poole and Rosenthal 1997). Roll-call votes came directly from the Brazilian
Chamber of Deputies Secretaria Geral da Mesa. The unit
of analysis in each case was the legislator-party, i.e., when,

79

PARTIES FOR RENT

a legislator cast votes as a member of more than one party,

I estimated separate ideal points for each party membership of that legislator.
I also included several time-specific variables to control for key moments and differences in the legislative
cycle. 50 th Leg is a dummy variable coded 1 for the
50th legislature and 0 for the 49th , capturing any change
in the overall switching rate and stabilization in the party
system. Campaign is a dummy variable coded 1 for municipal and national campaign months. Party-switching
is less likely during campaigns due to party membership
restrictions for candidates and the virtual shut-down of
the legislature during campaigns. Deadline is a dummy
variable, coded 1 for the last month when candidates
can change their party before registering to run for municipal or national office. Finally, Organize is a dummy
variable for the organizational period at the beginning of
each legislative session, prior to committee formation and
the election of Chamber leadership. This dummy variable
controls for the increased switching during that period as
coalitions jockey to maximize their leverage in committee
assignments and leadership votes.16 Finally, I included a
natural cubic spline to allow the hazard rate for switching to vary during the legislative session (Beck, Katz, and
Tucker 1999).
For Table 5, I simulated predicted probabilities for a
hypothetical legislator choosing between two parties with
all variables held constant across parties, except the factor
of interest. I report median simulated values and the .025
and .975 quantiles for the confidence intervals. For dis16
Note that all variables in this model are of two possible types.
Some variables simultaneously affect the decision to switch and
the choice of new party; other variables only affect the decision to
switch or not. A key distinction between the two is whether the variable is a characteristics of the party or the legislator. Characteristics
of parties affect both decisions to switch and destination choices.
These variables include Cabinet membership, size, Threshold, Ideology, and Major Party. Coefficients on such variables indicate how
the likelihood of joining a party would change if our explanatory
variable were increased in value.
Characteristics of individual legislators do not, in and of themselves, tell us much about where deputies might go. For example,
freshman should be more likely to switch than incumbents, but this
alone does not tell us where they might go. Similarly, knowing if a
legislator is a party leader or not helps us predict whether they will
switch, but tells us nothing about their likely destination.
This is a natural feature of the conditional logit, more fully discussed in Long (1997). In fact, in such models, individuals characteristics cannot be included directly, but only interacted with choice
characteristics. For this application, legislators characteristics in the
model were interacted with Home. A positive coefficient indicates
a greater likelihood of staying in ones current party; a negative
coefficient suggests the opposite. Education, as discussed, was also
interacted with Cabinet, Governor, and Distance to test additional
specific hypotheses.

crete variables, I calculated predicted probabilities for values as labeled in the table: For Governor, Gov=1, Opp=0.
For Threshold, Below=0, Above=1. For Cabinet, Opposition=0, Cabinet=1. For the continuous variables, I used
maximum and minimum values from the sample. Ideological distance was set to 1.4 (Far) and to 1.2 (Close).
Finally, Education was set to the maximum and minimum values observed in the dataset. Low Education was
14% of adults with four or more years of schooling; High
education was 82% of adults with four or more years of
schooling. I simulated confidence intervals by sampling
from estimates asymptotic distributions.

References

Abrucio,
F. 1998. Os Baroes da Federaca o: Os Governadores e a
Redemocratizaca o Brasileira. Sao Paulo: Editora Hucitec.
Aldrich, J. H. 1995. Why Parties? The Origin and Transformation
of Political Parties in America. Chicago: The University of
Chicago Press.
Aldrich, J. H., and W. T. Bianco. 1992. A Game-Theoretic
Model of Party Affiliation of Candidates and Office Holders. Mathematical Computer Modeling 16(89):10316.
Ames, B. 1995a. Electoral Rules, Constituency Pressures, and
Pork Barrel: Bases of Voting in the Brazilian Congress. Journal of Politics 57(2):32443.
Ames, B. 1995b. Electoral Strategy under Open-List Proportional Representation. American Journal of Political Science
39(2):40633.
Ames, B. 2001. The Deadlock of Democracy in Brazil. Ann Arbor:
The University of Michigan Press.
Amorim Neto, O. 1998. Of Presidents, Parties, and Ministers:
Cabinet Formation and Legislative Decision-Making Under
Separation of Powers. Ph.D. thesis. University of California,
San Diego.
Beck, N. L., J. Katz, and R. Tucker. 1999. Taking Time Seriously: Time-Series-Cross-Section Analysis with a Binary
Dependent Variable. American Journal of Political Science
42(4):126088.
Ben-Akiva, M., and S. R. Lerman. 1985. Discrete Choice Analysis:
Theory and Application to Travel Demand. Cambridge: The
MIT Press.
Bowler, S., D. M. Farrell, and R. S. Katz. 1999. Party Discipline
and Parliamentary Government. Columbus: The Ohio State
University Press.
Carey, J. M., and M. S. Shugart. 1995. Incentives to Cultivate
a Personal Vote: A Rank Ordering of Electoral Formulas.
Electoral Studies 14(4):41739.
Cox, G. W., and M. D. McCubbins. 1993. Legislative Leviathan.
Berkeley: University of California Press.
Cox, G. W., and F. M. Rosenbluth. 1995. Anatomy of a Split: the
Liberal Democrates of Japan. Electoral Studies 14(4):355
76.

80
Desposato, Scott. 2005. Conditional Logit with Unobserved
Choice Set Selecton, Working paper.
Glasgow, G. 2001. Mixed Logit Models for Multiparty Elections. Political Analysis 9(2):11636.
Graham, R. 1990. Patronage and Politics in Nineteenth-Century
Brazil. Stanford: Stanford University Press.
Haggard, S., and R. R. Kaufman. 1995. The Political Economy
of Democratic Transitions. Princeton: Princeton University
Press.
Heller, W. B., and Mershon C. 2005. Party Switching in the
Italian Chamber of Deputies, 19962001.Journal of Politics
67(2):53659.
Johnson, G., and B. Crisp. 2003. Mandates, Powers, and Policies. American Journal of Political Science 47(1):12842.
Keck, M. 1995. The Workers Party and Democratization in Brazil.
New Haven: Yale University Press.
Leal, V. N. 1977. Coronelismo: The Municipality and Representative Government in Brazil. New York: Cambridge University
Press.
Long, J. S. 1997. Regression Models for Categorical and Limited
Dependent Variables. Number 7 in Advanced Quantititative
Techniques in the Social Sciences Series. Thousand Oaks:
Sage Publications.
Long, J. S., and J. Freese. 2001. Regression Models for Categorical
Dependent Variables Using Stata. College Station, TX: Stata
Press.
Maddala, G. S. 1983. Limited-Dependent and Qualitative Variables in Econometrics. Cambridge: Cambridge University
Press.
Mainwaring, S. 1997. Mutipartism, Robust Federalism, and
Presidentialism in Brazil. In Presidentialism and Democracy
in Latin America, eds. S. Mainwaring and M. S. Shugart. New
York: Cambridge University Press, pp. 55109.
Mainwaring, S., and A. P. Lina n 1997. Party Discipline in the
Brazilian Constitutional Congress. Legislative Studies Quarterly 12(4):45383.
Mainwaring, S. P. 1999. Rethinking Party Systems in the Third
Wave of Democratization: The Case of Brazil. Stanford: Stanford University Press.
McFadden, D. 1973. Conditional Logit Analysis of Qualitative
Choice Behavior. In Frontiers in Econometrics, ed. P. Zarembka. New York: Academic, pp. 10542.

SCOTT W. DESPOSATO

McLaughlin, K. J. 1991. A Theory of Quits and Layoffs with

Efficient Turnover. Journal of Political Economy 99(1):129.
Mejia Acosta, A. 2004. Ghost Coalitions: Economic Reforms, Fragmented Legislatures and Informal Institutions in Ecuador.
Ph.D. thesis. University of Notre Dame.
Melo, C. R. F. d. 2000. Partidos e Migracao Partidaria na Camara
dos Deputados. Dados 43(2).
Mershon, C., and W. B. Heller. 2004. Theoretical and Empirical
Models of Party Switching. Memo Prepared for the Dublin
Meeting of the Research Work Group on Legislative Party
Switching, Trinity College, 48 July 2004.
National Democratic Institute. 1997. Strengthening Nepals
Multiparty Democracy: Party Discipline and Anti-Defection
Measures. A National Democratic Institute for International Affairs Workshop.
Neto, O. A. 2002. Parties and Cabinets Dataset.
Nicolau, J. M. 1996. A migracao partidaria na Camara dos Deputados (19911996). Monitor Publico 3(10):4145.
Pereira, C., and B. Mueller. 2004. The Cost of Governing:
Strategic Behavior of the President and Legislators in Brazils
Budgetary Process. Comparative Political Studies 20(10):1
32.
Poole, K. T., and H. Rosenthal. 1997. Congress: A PoliticalEconomic History of Roll Call Voting. New York: Oxford University Press.
Reed, S. R., and E. Scheiner. 2002. Electoral Incentives and
Policy Preferences: Mixed Motives Behind Party Defections
in Japan. British Journal of Political Science 33(3):46990.
Samuels, D. J. 2003. Ambition, Federalism, and Legislative
Politics in Brazil. New York: Cambridge University Press.
Samuels, D. J. (Forthcoming). Sources of Mass Partisanship in
Brazil. Latin American Politics and Society.
Sartori, G. 1993. Nem Presidencialismo, Nem Parlamentarismo. Novos Estudos CEBRAP 35:314.
Schmitt, R. 1999. Migracao partidaria e reeleicao na Camara
dos Deputados. Novos Estudos CEBRAP 54:12746.
Schneider, A. 2001. Federalism against Markets: Local Struggles
for Power and National Fiscal Adjustment in Brazil. Ph.D.
thesis. University of California, Berkeley.
Snyder Jr., J. M., and M. M. Ting. 2001. An Informational
Rationale for Political Parties. American Journal of Political
Science 46(1):90110.